To be able to claim a world record, one must compute Pi using two different methods and (hopefully) get the same answer on both runs. One run can be viewed as the computation and the other as the verification, although the situation is completely symmetrical. Both computations were performed in base-10 throughout since the conversion from base-2 to base-10 would be very expensive.
| HITACHI SR8000 | theoretical speed | memory used |
| 1 node | 8 Gflop/s | 6.758 GB |
| all 128 nodes | 1 Tflop/s | 865 GB |
| run | algorithm | time taken (h:m:s) |
| 1. | Gauss-Legendre algorithm | 37:21:04 |
| 2. | Borwein's 4-th order convergent algorithm | 46:07:10 |
According to Stu's pi page, the fastest PC program to compute Pi is PiFast. By interpolating the time taken by PiFast to compute 1 million and 64 million digits in core on a Pentium 450MHz, I get the following formula:
By extrapolating, I estimate that computing 206 billion digits should take 36000 hours (4.1 years) on the same Pentium assuming an unbounded main memory. The supercomputer run was about 1000 times faster than that. This is enough to convince me that the computation was efficiently carried.

based on this Brief History of Pi Calculation with Computers